Machines

To address the issue of reduced fault diagnosis accuracy caused by insufficient samples in laboratory datasets, this study proposes an improved Transfer Component Analysis (TCA) algorithm with dynamic kernel parameter adjustment, combined with Local Mean Decomposition (LMD). Firstly, the original signals are decomposed using LMD, and representative signal components are reconstructed based on the Pearson’s correlation coefficient to enhance feature representativeness. Then, multidimensional features, including Root Mean Square (RMS), kurtosis, and main frequency (MF), are extracted from the reconstructed signals to comprehensively reflect signal characteristics in terms of energy distribution, impact properties, and frequency structure. Subsequently, a dynamic kernel parameter adjustment strategy is incorporated into TCA to adaptively optimize the kernel parameters, effectively reducing the distribution discrepancy between the source and target domains and enhancing the generalization capability of cross-domain feature transfer. Finally, a Least Squares Support Vector Machine (LSSVM) classifier is employed to perform fault diagnosis on the reconstructed features. The experimental results demonstrate that the proposed method achieves significantly higher diagnostic accuracy than traditional approaches under various operating conditions, especially when signals are complex and distribution differences are large, showing strong robustness and adaptability.

Rolling bearings are crucial components in CNC machine tool spindles, and their health condition directly affects machining precision and operational reliability. To address the significant challenges of bearing fault diagnosis in industrial environments, this paper proposes an adaptive shapelet-based deep learning model for bearing fault diagnosis. The proposed model integrates three key components: (1) an adaptive multi-scale shapelet extraction module for discriminative pattern learning, (2) a gated parallel CNN with depthwise separable convolutions for multi-scale spatial feature extraction, (3) an enhanced bidirectional long short-term memory network with residual connections for temporal dependency modeling. A composite loss function combining cross-entropy, supervised contrastive learning, and multi-scale consistency regularization is employed for training. To simulate real-world industrial noise conditions, Gaussian, uniform, and impulse noise were injected into the signals. Experiments conducted on the CWRU and IMS datasets demonstrate that, compared with state-of-the-art methods, the proposed approach achieves stronger noise robustness, higher fault classification accuracy, and more stable performance under severe noise contamination.

The three-level neutral-point-clamped cascaded rectifier (3LNPC-CR) is a key component in power electronic transformers (PET) due to its high efficiency and modular configuration. However, voltage imbalance among submodule DC links may cause system instability and degrade power quality. To address this issue, this paper proposes a voltage balancing strategy based on Model Predictive Control with a Smooth Switching Sequence (MPC-3S). First, a negative-sequence current control strategy is introduced to equalize the voltages among phases. In addition, an improved modulation scheme is developed to predict and optimize system states in real time within the control horizon, dynamically selecting the optimal switching sequence to achieve rapid voltage equalization. Finally, simulation and experimental results on a three-phase, three-module 3LNPC-CR prototype demonstrate that the proposed MPC-3S strategy can achieve fast intra-phase voltage balancing within 0.1 s under load imbalance, while maintaining high-quality grid-side current. These results verify that the proposed method significantly enhances both the dynamic and steady-state performance of 3LNPC-CR systems, providing a practical and efficient solution to the voltage-balancing challenge in PET applications.